Re: [PATCH] Split gcc.dg/builtins-20.c

[Andrew Pinski <pinskia at physics dot uc dot edu>]

I would split it up even further moving out *l of the *f tests.
Then I would xfails on non linux, not just execute it only on linux.

Thanks,
Andrew Pinski


On Friday, Jun 13, 2003, at 23:17 US/Eastern, Roger Sayle wrote:

> My recent test case builtins-20.c is currently failing on sh-elf
> and presumably several other non-linux platforms (thanks to Joern
> Rennecke for bringing this to my attention).  The problem is that
> this optimization/test case attempt to synthesize the use of a
> C99 function that isn't guaranteed to be available on all platforms.
>
> So although "sin(x)/cos(x) != tan(x)" should work everywhere, the
> optimizer will only convert "sinf(x)/cosf(x)" into tanf(x) on platforms
> where it is known this function will be present.  So this testcase
> has been failing on those platforms where GCC can't guarantee a c99
> library.
>
> The following patch splits the original testcase gcc.dg/builtins-20.c
> into two parts.  One test program for double functions that only
> requires c90 functionality remains in builtins-20.c, and a second
> test program builtins-22.c that contains the float and long double
> tests, which is only executed on *-*-linux*.
>
>
> [In practice, the test2f and test2l optimizations will be applied
> everywhere, but the split below is much cleaner conceptually.]
>
> Ok for mainline?